68020 bitfield instructions in winuae

[rsn8887]

I am trying to find the 68020 bitfield instruction emulation in newcpu.cpp. Any hints where I might find those?

I am trying to backport a fix for guru in Virocop AGA in UAE4All2 Vita and I have a hunch it might be a problem with the bitfield instructions.

[Toni Wilen]

All instruction emulation functions are located in CPU specific generated cpuemu_xx files + bitfield fetch/store helpers in newcpu_common.

[rsn8887]

Ah thanks for the hint. It looks much more complicated than I hoped.

[Toni Wilen]

Yes, it isn't that simple but it shouldn't be too complex, first check bitfield instruction generation code in gencpu.cpp.

[rsn8887]

Since uae4all vita uses Famec I have to compare quite different sources.

Was there maybe an older version of WinUAE (from years ago) on github, where the bitfield operations where not yet auto-generated? It might make debugging uae4all vita easier if I could just update the ops one by one and see which one fixes the crash.

[Toni Wilen]

WinUAE never had non-generated 68k emulation. It was some really early UAE that did it differently, before "newcpu" existed.

I remember there was some bitfield fix ages ago, I think something in Dungeon Master II didn't work without the fix.

[rsn8887]

Thanks! I found the fix you mentioned:
http://eab.abime.net/showpost.php?p=15755&postcount=23

I am looking at these bitfield instructions in the latest UAE4All code, where famec had been fixed by TomB a couple times already:
https://github.com/rsn8887/uae4all2/...codes.h#L45673

I see a lot of commented out "& 7" parts, like:
DREG((res >> 12) /*& 7*/) = dst;

as well as a suspicious comment in BFCHG

// BFCHG Dn - 68020+
OPCODE(0xEAC0)
{
   u32 res;
   u32 dst;
   u32 mask, offset, width;
   
   FETCH_WORD(res)
   BF_REG_GET(res, dst, offset, width)
   mask = ((u32)0xFFFFFFFF << (32 - width)) >> offset;
//   BF_MASK(mask, offset, width)
//   BF_SHIFT_UP(mask, offset, width)
   DREG((Opcode /*>> 0*/) & 7) ^= mask;
	RET(12)
}

in BFCLR:

// BFCLR Dn - 68020+
OPCODE(0xECC0)
{
   u32 adr, res;
   u32 dst;
   u32 mask, offset, width;
   
   FETCH_WORD(res)
   BF_REG_GET(res, dst, offset, width)
   mask = ((u32)0xFFFFFFFF << (32 - width)) >> offset;
//   BF_MASK(mask, offset, width)
//   BF_SHIFT_UP(mask, offset, width)
   DREG((Opcode /*>> 0*/) & 7) &= ~mask;
	RET(12)
}

and an &=~ repeatedly in BFCLR:

// BFCLR (An) - 68020+
OPCODE(0xECD0)
{
   u32 adr, res;
   u32 dst;
   u32 mask, width;
   s32 offset;
   u32 bf0, bf1;

   FETCH_WORD(res)
   adr = AREG((Opcode /*>> 0*/) & 7);
   BF_GET_PARM(res, offset, width)
   BF_MEM_GET(&adr, &dst, &offset, width, &bf0, &bf1);
   BF_SET_FLAGS(dst, width)
   BF_MASK(mask, offset, width)
//   dst &= ~mask;
   BF_MEM_PUT(adr, dst, 0/*mask*/, offset, width, bf0, bf1);
	RET(20)
}

Does any of this this look faulty to you?

[rsn8887]

Hmm it seems like the comments make sense if I assume someone improved the functions and commented out the old pieces instead of deleting them.

Now I am trying to compare the famec code in UAE4All2 here
https://github.com/rsn8887/uae4all2/...codes.h#L45673

with Toni's code here
https://github.com/tonioni/WinUAE/bl...ncpu.cpp#L5188

It would be good to know which exact instruction ViroCop AGA crashes at. I will try to use the WHDLoad expert switch to get more information.

[rsn8887]

I ran WHDLoad Expert and now I am not so sure it is the bitfield operations anymore since the pc stack just shows a bunch of regular btst operations and no bitfield op.

I am attaching a screenshot of the dump. Toni if you have a second maybe you can glance at this and see if it means anything to you?

Does this reveal the bug in the emulation?

Edit: I launched a few more times again and those times I got a line1111 error:

[rsn8887]

I guess the error happened before and caused some erroneous jump because the instruction sequence shown by WHDLoad makes no sense.

[rsn8887]

Update: I found a bitfield instruction that is executed just before the crash (Guru meditation) in Virocop AGA on UAE4All2 Vita/Switch:
https://github.com/rsn8887/uae4all2/...codes.h#L46347

it is this instruction:

// BFFFO Dn - 68020+
OPCODE(0xEDC0)
{
u32 adr, res;
u32 src, dst;
u32 mask, offset, width;

FETCH_WORD(res)
BF_REG_GET(res, src, offset, width)
BF_FFO(src, mask, offset, width)
RET(18)
}

I think the implementation of this might be faulty?

@ToniWilen: Can you see anything wrong with it?

EDIT: I think I need to implement this patch: http://eab.abime.net/showpost.php?p=1051962&postcount=9

[rsn8887]

That fixed it! No more crashes in ViroCop AGA. I suspect it is the only game ever that uses these bitfield instructions, though.

[rsn8887]

@Toni Wilen: Does my fix make sense to you? If you have a second, it is here:
https://github.com/rsn8887/uae4all2/...15d816d1e7ee76

[Toni Wilen]

They look totally different compared to (win)uae implementation. I am not going to try to understand how it works in this port, sorry.

[rsn8887]

That’s fine thank you for taking a look. I went through them step by step to ensure they are working the same way as in WinUAE, so it should be ok.